CIM-Embed: Instruction-Conditioned Multimodal Embeddings for Clustering

CIM-Embed enables multimodal clustering by learning embeddings conditioned on specific instructions, such as grouping products by color or material. It uses supervised contrastive learning to map images, text, and instructions into a single space for use with HDBSCAN.